Implement Support for no_std #341

CryZe · 2019-09-07T10:16:50Z

This adds a new std feature to chrono that is enabled by default. By deactivating this feature via default-features = false you can now use chrono in applications that don't use the standard library. The serde feature is supported as well.

Resolves #336

This adds a new `std` feature to chrono that is enabled by default. By deactivating this feature via `default-features = false` you can now use chrono in applications that don't use the standard library. The `serde` feature is supported as well. Resolves chronotope#336

CryZe · 2019-09-07T10:32:29Z

Well looks like I definitely need help with that CI script.

quodlibetor · 2019-09-07T15:27:34Z

Thank you! I will look at the CI script.

quodlibetor

I've got a couple nits that I'd be happy to take care of if you don't feel like it, otherwise this seems great!

ci/travis.sh

quodlibetor · 2019-09-07T15:47:05Z

src/datetime.rs

 use std::time::{SystemTime, UNIX_EPOCH};
 use oldtime::Duration as OldDuration;

+#[cfg(not(any(feature = "std", test)))]
+use alloc::string::{String, ToString};


It seems like it might be worth sticking this (and what seems like the two functions that return strings) behind a separate alloc feature flag?

I added an alloc feature now. I'm kind of sad how much is missing without it though. And it's mostly only because the timezone offset is stored as a String in the DelayedFormat.

CryZe · 2019-09-09T11:47:23Z

src/format/mod.rs

+mod core_only {
+    /// Core only
+    #[derive(Debug, PartialEq, Eq, PartialOrd, Ord)]
+    pub struct Box<T: ?Sized>(core::marker::PhantomData<T>);


This thing is kind of hacky, but it allows at least partially supporting some of the format module stuff.

Can you explain how/if this actually works/what the runtime experience would be? Or is there some explanation of this pattern somewhere?

Alternatively, would you mind adding some small code to the core test file that demonstrates what this ends up doing?

It's only so the Item enum actually compiles as there's Box<str> in there. I'm not sure if we even need this enum when alloc isn't compiled in. I honestly don't know what even works in that module at all if DelayedFormat isn't compiled in.

Does the timezone for the DelayedFormat really need to be a String or could it be turned into a &'static str? (I'm not sure if there ever are any truly dynamic timezones that would need String)

Wait, what is OwnedLiteral and OwnedSpace even used for? I don't see them being used anywhere.

Yes I ended up putting these variants behind the same feature gate, I'm not sure that it will have any effect on usability. Potentially if people were compiling their own format::Item lists it could have been a thing?

CryZe · 2019-09-09T11:49:53Z

Also with alloc this is now slightly a breaking change as the serde feature is now called serde-1, because you can't have feature dependencies on optional crates themselves (for whatever reason).

Although it requires a semver incompatible version bump anyway, as adding new features such as std and alloc would be incompatible with someone using chrono with default-features = false.

To ensure that we don't accidentaly not verify that chrono compiles for core.

quodlibetor · 2019-09-13T18:51:28Z

Also.. this is now slightly a breaking change

Hm that's a good point, I'll need to think about that. Does SemVer apply to compilation flags? I've taken things out of the --no-default-features install without bumping the patch version before and I'm not sure that I've gotten bug reports about it. This is a much more dramatic change than before, though, so maybe not.

as the serde feature is now called serde-1, because you can't have feature dependencies on optional crates themselves.

Yeah that's true, but you can rely on optional dependencies as though they're features, so I think that it's fine to just use serde, and it seems to work for me.

CryZe · 2019-09-13T19:38:17Z

I believe if you activate the serde feature without the alloc feature now, you get weird compilation errors. I guess if it's properly documented that you need to also activate alloc then, it might not be a problem.

Or you could even do something like:

#[cfg(all(feature = "serde", not(feature = "alloc"))]
compile_error!("For the serde feature you also need the alloc feature.");

(this is not the full cfg, as std and test are also valid)

quodlibetor · 2019-09-13T19:59:11Z

Yeah that's a good point, I didn't quite realize that it was possible for people to want to be able to use serde without std, but in retrospect it makes sense.

quodlibetor · 2019-09-13T21:04:44Z

Just to be clear, you mean that serde requires the alloc feature, so doing --no-default-features --features serde that will start failing, but I would expect just doing --features serde would work. And that's what appears to be the case, but I might be missing something.

CryZe · 2019-09-13T21:06:20Z

I'd be surprised if it works, but I guess it might? I'll check tomorrow.

Slightly easier to reason about the code via some code movement, printing some banners to make it more obvious when cargo is being run since it is run so many times.

Part of this means that we don't actually need to allocate to write to Serde any more, which is a win in its own right.

quodlibetor · 2019-09-16T00:50:27Z

I just pushed a change that should remove all intermediary allocation even when the serde feature is enabled. I didn't look into how that plays with the core::box placeholder.

I tried to simplify the features, as well, let me know how it feels to you to use this way.

CryZe · 2019-09-20T13:28:53Z

That does not seem to build if you build with serde and no alloc:

   Compiling chrono v0.4.9 (https://github.com/CryZe/chrono?branch=no-std#e5bbc94c)
error[E0432]: unresolved import `alloc`
    --> /home/user/.cargo/git/checkouts/chrono-5169c63406b50cd5/e5bbc94/src/naive/datetime.rs:1761:13
     |
1761 |         use alloc::format;
     |             ^^^^^
     |             |
     |             unresolved import
     |             help: a similar path exists: `core::alloc`

error[E0432]: unresolved import `alloc`
    --> /home/user/.cargo/git/checkouts/chrono-5169c63406b50cd5/e5bbc94/src/naive/datetime.rs:1908:13
     |
1908 |         use alloc::format;
     |             ^^^^^
     |             |
     |             unresolved import
     |             help: a similar path exists: `core::alloc`

error[E0432]: unresolved import `alloc`
    --> /home/user/.cargo/git/checkouts/chrono-5169c63406b50cd5/e5bbc94/src/naive/datetime.rs:2055:13
     |
2055 |         use alloc::format;
     |             ^^^^^
     |             |
     |             unresolved import
     |             help: a similar path exists: `core::alloc`

error: cannot find macro `format` in this scope
    --> /home/user/.cargo/git/checkouts/chrono-5169c63406b50cd5/e5bbc94/src/naive/time.rs:1448:51
     |
1448 |             value.parse().map_err(|err| E::custom(format!("{}", err)))
     |                                                   ^^^^^^

error: cannot find macro `format` in this scope
    --> /home/user/.cargo/git/checkouts/chrono-5169c63406b50cd5/e5bbc94/src/naive/datetime.rs:1711:51
     |
1711 |             value.parse().map_err(|err| E::custom(format!("{}", err)))
     |                                                   ^^^^^^

error: cannot determine resolution for the macro `format`
    --> /home/user/.cargo/git/checkouts/chrono-5169c63406b50cd5/e5bbc94/src/naive/datetime.rs:1857:46
     |
1857 |                     .ok_or_else(|| E::custom(format!("value is not a legal timestamp: {}", value)))
     |                                              ^^^^^^
     |
     = note: import resolution is stuck, try simplifying macro imports

error: cannot determine resolution for the macro `format`
    --> /home/user/.cargo/git/checkouts/chrono-5169c63406b50cd5/e5bbc94/src/naive/datetime.rs:1865:46
     |
1865 |                     .ok_or_else(|| E::custom(format!("value is not a legal timestamp: {}", value)))
     |                                              ^^^^^^
     |
     = note: import resolution is stuck, try simplifying macro imports

error: cannot determine resolution for the macro `format`
    --> /home/user/.cargo/git/checkouts/chrono-5169c63406b50cd5/e5bbc94/src/naive/datetime.rs:2004:46
     |
2004 |                     .ok_or_else(|| E::custom(format!("value is not a legal timestamp: {}", value)))
     |                                              ^^^^^^
     |
     = note: import resolution is stuck, try simplifying macro imports

error: cannot determine resolution for the macro `format`
    --> /home/user/.cargo/git/checkouts/chrono-5169c63406b50cd5/e5bbc94/src/naive/datetime.rs:2012:46
     |
2012 |                     .ok_or_else(|| E::custom(format!("value is not a legal timestamp: {}", value)))
     |                                              ^^^^^^
     |
     = note: import resolution is stuck, try simplifying macro imports

error: cannot determine resolution for the macro `format`
    --> /home/user/.cargo/git/checkouts/chrono-5169c63406b50cd5/e5bbc94/src/naive/datetime.rs:2150:46
     |
2150 |                     .ok_or_else(|| E::custom(format!("value is not a legal timestamp: {}", value)))
     |                                              ^^^^^^
     |
     = note: import resolution is stuck, try simplifying macro imports

error: cannot determine resolution for the macro `format`
    --> /home/user/.cargo/git/checkouts/chrono-5169c63406b50cd5/e5bbc94/src/naive/datetime.rs:2157:46
     |
2157 |                     .ok_or_else(|| E::custom(format!("value is not a legal timestamp: {}", value)))
     |                                              ^^^^^^
     |
     = note: import resolution is stuck, try simplifying macro imports

error: aborting due to 11 previous errors

quodlibetor · 2019-09-20T13:50:02Z

Oh, I thought i had fixed that correctly, I'll get it working.

They just add noise, now.

quodlibetor · 2019-09-20T17:50:52Z

This seems like it's working on stable, now. The failures had to do with using core in 1.13, which I'm not super interested in debugging unless someone actively wants it.

CryZe · 2019-09-23T09:41:19Z

Yeah it seems to be working now across all the different combinations :)

quodlibetor · 2019-09-23T16:50:48Z

WooHoo!

Okay I'd like to see what we can do about the stub core module that you created, ideally we can make it unnecessary. I doubt I'll have any time to look at it before friday, if you're interested.

CryZe · 2019-10-30T16:01:12Z

Do you still want me to check it? I don't really understand the formatting module enough, but I could take a closer look and try to understand it if you don't have time to look into it yourself. I'd like to get this merged (and released) fairly soon, as the other crates that are blocking me almost all have released new versions and the few remaining are intending to do so within the next weeks.

This means that a few more features of formatting items don't compile in non-alloc environments, but they wouldn't have worked correctly anyway.

quodlibetor · 2019-11-22T21:53:06Z

I am happy with this as is now, we can add and improve things as they come up. Thanks for this!

CryZe added 2 commits September 7, 2019 12:12

Use std in all other tests on CI

5ccec02

CryZe force-pushed the no-std branch from 2b9626e to 5ccec02 Compare September 7, 2019 10:24

CryZe mentioned this pull request Sep 7, 2019

Support no_std use cases by using alloc LiveSplit/livesplit-core#223

Closed

26 tasks

quodlibetor self-assigned this Sep 7, 2019

quodlibetor reviewed Sep 7, 2019

View reviewed changes

CryZe force-pushed the no-std branch from 7f10269 to bab2b00 Compare September 9, 2019 11:44

CryZe commented Sep 9, 2019

View reviewed changes

Introduce an alloc feature

e62a054

CryZe force-pushed the no-std branch from bab2b00 to e62a054 Compare September 9, 2019 11:51

quodlibetor added 2 commits September 13, 2019 14:49

Rename serde-1 back to serde

4027bbb

Add a trivial function in core test

5e1e2d5

To ensure that we don't accidentaly not verify that chrono compiles for core.

Make CI able to run non-cargo commands in channel

e8c708d

quodlibetor force-pushed the no-std branch from e6ac38b to e8c708d Compare September 13, 2019 19:50

quodlibetor added 2 commits September 15, 2019 20:44

Make travis.sh and Makefile "nicer"

5b72ef3

Slightly easier to reason about the code via some code movement, printing some banners to make it more obvious when cargo is being run since it is run so many times.

First pass at making "alloc" its own feature

e5bbc94

Part of this means that we don't actually need to allocate to write to Serde any more, which is a win in its own right.

quodlibetor added 2 commits September 20, 2019 10:50

Test against serde with no features in both std and no-std

a09f9ba

put bincode back to the version used for testing

ffcd0f9

quodlibetor force-pushed the no-std branch from 2b3ec4a to ffcd0f9 Compare September 20, 2019 15:02

quodlibetor added 4 commits September 20, 2019 11:19

Make ci script more understandable and robust, hopefully

9a3e489

support rust v1.13.0 struct initialization syntax

911dc57

Using core-only on 1.13 doesn't seem to work the same as stable

6499c5b

Remove some pwds from the CI script

918cff1

They just add noise, now.

Remove core_only, cfg-out the format::Item::Owned* variants

64a28d6

This means that a few more features of formatting items don't compile in non-alloc environments, but they wouldn't have worked correctly anyway.

quodlibetor merged commit d9929a6 into chronotope:master Nov 22, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implement Support for no_std #341

Implement Support for no_std #341

CryZe commented Sep 7, 2019

CryZe commented Sep 7, 2019

quodlibetor commented Sep 7, 2019

quodlibetor left a comment

quodlibetor Sep 7, 2019

CryZe Sep 9, 2019

CryZe Sep 9, 2019

quodlibetor Sep 13, 2019

CryZe Sep 13, 2019 •

edited

CryZe Sep 13, 2019

CryZe Sep 13, 2019

quodlibetor Nov 23, 2019

CryZe commented Sep 9, 2019 •

edited

quodlibetor commented Sep 13, 2019

CryZe commented Sep 13, 2019 •

edited

quodlibetor commented Sep 13, 2019

quodlibetor commented Sep 13, 2019

CryZe commented Sep 13, 2019

quodlibetor commented Sep 16, 2019

CryZe commented Sep 20, 2019 •

edited

quodlibetor commented Sep 20, 2019

quodlibetor commented Sep 20, 2019

CryZe commented Sep 23, 2019

quodlibetor commented Sep 23, 2019

CryZe commented Oct 30, 2019

quodlibetor commented Nov 22, 2019

Implement Support for no_std #341

Implement Support for no_std #341

Conversation

CryZe commented Sep 7, 2019

CryZe commented Sep 7, 2019

quodlibetor commented Sep 7, 2019

quodlibetor left a comment

Choose a reason for hiding this comment

quodlibetor Sep 7, 2019

Choose a reason for hiding this comment

CryZe Sep 9, 2019

Choose a reason for hiding this comment

CryZe Sep 9, 2019

Choose a reason for hiding this comment

quodlibetor Sep 13, 2019

Choose a reason for hiding this comment

CryZe Sep 13, 2019 • edited

Choose a reason for hiding this comment

CryZe Sep 13, 2019

Choose a reason for hiding this comment

CryZe Sep 13, 2019

Choose a reason for hiding this comment

quodlibetor Nov 23, 2019

Choose a reason for hiding this comment

CryZe commented Sep 9, 2019 • edited

quodlibetor commented Sep 13, 2019

CryZe commented Sep 13, 2019 • edited

quodlibetor commented Sep 13, 2019

quodlibetor commented Sep 13, 2019

CryZe commented Sep 13, 2019

quodlibetor commented Sep 16, 2019

CryZe commented Sep 20, 2019 • edited

quodlibetor commented Sep 20, 2019

quodlibetor commented Sep 20, 2019

CryZe commented Sep 23, 2019

quodlibetor commented Sep 23, 2019

CryZe commented Oct 30, 2019

quodlibetor commented Nov 22, 2019

CryZe Sep 13, 2019 •

edited

CryZe commented Sep 9, 2019 •

edited

CryZe commented Sep 13, 2019 •

edited

CryZe commented Sep 20, 2019 •

edited